A good user experience is the key to reach the end goal: conversational AI success, according to a new report from Forrester.
The concept of chatbots is heralded, and for good reason--it can create faster self-service features for customers and employees. Additionally, chatbots provide a sense of immediacy, modernity and customer-first. They're also economically sound, when compared to the alternative of live on-call help desks. Unfortunately, as a new report from Forrester reveals, chatbots repeatedly garner low-satisfaction scores. Great idea, bad reputation.
Certainly the first step company leaders need to take is to find skilled IT professionals to develop a customized, worry-free system and, in its report, Forrester presented "Five strategic principles to help infrastructure and operations professionals ensure success and avoid common chatbot pitfalls."
SEE: TechRepublic Premium editorial calendar: IT policies, checklists, toolkits, and research for download (TechRepublic Premium)
Streamlining customer service
Chatbots provide organizations with a new and automated channel for request fulfillment. In last year's Forrester Analytics survey, 58% of surveyed global IT decision-makers report adopting or having adopted chatbots. In 2018, 54% of U.S. online consumers responded negatively to chatbots in a Forrester survey.
Five strategic principles
In order to find success, companies "must keep their users at the center of their help desk chatbot strategies and embrace five principles."
1. Build a chatbot team with cross-organization input
- Most organizations build chatbots for multiple departments, use cases and functions, thus decentralizing efforts. Forrester said that the first step in adopting an effective chatbot strategy is having someone (or better yet, a team), to oversee chatbot operations who reports to the chief information officer.
- Design the chatbot platform team with a hub-and-spoke model: The model involves a central chatbot platform team acting as the hub and subject/function experts, such as help desk agents, as spokes. It unifies strategy, meets the needs of end users and is extensible as use-cases grow. The chatbot team can also be scaled up or down, depending on need.
- Get honest, detailed user feedback. Get it early and apply it soonest.
- Work with help desk agents directly to match the initial requests with workflows. This accelerates identification of high-volume requests, builds templates for automation and rapidly optimizes language and intent for the enterprise's internal vocabulary.
2. Keep user experience a top priority
- Don't limit chatbot strategy to implementation and outcome. User experience is often ignored and the platform will fail. Apply feedback.
- Use visuals to show what the chatbot does. Display the most common requests that the chatbot can handle, instead of just, "How can I help you today?," have the introduction ask "Do you need help with…" with common issues. If it can't do something yet, say so, and then get a human.
- Keep humans in the loop. Give users the option to talk to a human, even as a callback and automatically transfer to a human agent on the third failed attempt, which ensures some resolution.
- Avoid page pushing. There's a tendency for chatbots to answer questions by directing users to hyperlinks or forms, aka page pushing. Bring only immediately relevant information, not the entire page, into the chat window. Chatbot solutions now have page-excerpting, knowledge-extraction and form-indexing capabilities to show relevant information within the chatbot.
3. Pick tools based on existing expertise, and stick to OOTB features
- Budget accordingly. "Tool investment presents an opportunity to optimize effort, design, and spend efficiency," Forrester reported. Make the most of your investments with tool selection practices.
- Pick tools that support your workforce's existing skills. Finding the right balance of preexisting and custom-built options depends on the workforce. Chatbot vendors can recommend development and support staffing to help guide decisions.
- Use OOTB capabilities. Chatbots are hard, first impressions are crucial and chatbot development is difficult and time-intensive. Use OOTB everything wherever possible. These OOTB features can be modules that help decipher "chitchat" (i.e., general conversational vocabulary), industry-specific language, canned integrations or fully vendor-managed chatbots that come pretrained on a given workplace. Resist customizing, consistently optimize chatbots.
- Use tools that leverage real data. Instead of developing a chatbot based on perceived user questions, choose platforms that develop it from real user questions. Resolved tickets and agent-user interactions with high satisfaction scores can rapidly inform your chatbots and are critical data to process.
4. Plan a practical capability rollout
- Balance operations and automation best practices. Chatbots are best suited to automate low-variability, high-frequency requests. Start with the top 10 requests and establish a clear minimum viable functionality strategy prior to release. Maintenance should focus on continuous language improvement. To best serve challenging requests, the help desk chatbot must appear helpful.
- Reduce upfront work for agents. Agent automation features, such as similar incident identification, pre-agent interaction classification, automated data collection and faster human-handled requests, which provides a system for ongoing improvement.
- Prioritize continuous improvement of capabilities. Execute rapid iterations and improvements to existing functions. Focus on improvement, not new features.
- Prioritize completing a few tasks at a high success rate. Completing tasks at a high sucess rate is preferable to multiple features at a low success rate.
5. Measure what really matters: users' success
- Chatbot success means reduced support costs and significant return on investment. Failure results in poor support experiences and users flocking to more expensive alternative mechanisms, ballooning costs.
- Measure balanced interaction success metrics. Don't just measure success/failure rates.
- Measure perception to identify overall success. Solicit feedback from end users. Forrester recommends using a total outcome score to measure the overall effectiveness of your service.
- Eliminate key performance indicators that discourage automation.
Meeting the metrics
Chatbot interaction metrics matter and Forrester presents the metric and target:
- Intent recognition rate: More than 90% "is a good target"
- Dropped conversations: Lower is better, measure weekly, keep in low percentages
- Number of interactions per user: A high number of chatbot interactions; can also advise company of issues based on repeated support requests
- User mean time to recovery: Measure MTTR in hours, a lower number is best; measure automated ticket resolutions in minutes
- Chatbot ticket deflection rate: For solutions using fewer out-of-the-box features, 20% deflection of tickets/month should be the six-month target; for more OOTB features, 40% to 50% deflection is achievable
- Total outcome score: Measure TOS weekly, a high percentage of users respond "yes"
- IT leader's guide to deep learning (TechRepublic Premium)
- Building the bionic brain (free PDF) (TechRepublic)
- Hiring Kit: Autonomous Systems Engineer (TechRepublic Premium)
- What is AI? Everything you need to know about Artificial Intelligence (ZDNet)
- Artificial Intelligence: More must-read coverage (TechRepublic on Flipboard)